sglang
https://github.com/sgl-project/sglang
Python
SGLang is a high-performance serving framework for large language models and multimodal models.
Triage Issues!
When you volunteer to triage issues, you'll receive an email each day with a link to an open issue that needs help in this project. You'll also receive instructions on how to triage issues.
Triage Docs!
Receive a documented method or class from your favorite GitHub repos in your inbox every day. If you're really pro, receive undocumented methods or classes and supercharge your commit history.
Python not yet supported1 Subscribers
Add a CodeTriage badge to sglang
Help out
- Issues
- [Fix] Preserve Chat/Responses logprob candidates and empty-delta records
- [Fix] Correct FlashKDA prefill return contract with regression tests
- [PD disaggregation] Duplicate client-supplied `bootstrap_room` crashes prefill and decode schedulers (uncaught `KeyError` / `ValueError`) # Checklist
- [Bug] [diffusion] MiniMax-H3 INT8 ConvRot checkpoints skip the head-interleaved QKV reorder, silently corrupting video/audio (same defect as #34227, non-FSDP path)
- chore: bump docs install version to 0.5.20
- [Bugfix] Accept pp_proxy_tensors on remaining spec-v2 workers
- [Spec] Honor the DFlash anchor-first draft block layout
- [Bug] glm5_next: MTP layer (layers.45) not skipped for 'model.language_model.*' names — local-inference-lab/GLM-5.3-Flash-NVFP4 (MXFP8 MTP experts) fails to load with a w2 shape mismatch
- [Bug] GLM-5.3-Flash has no usable DSA attention backend on SM121 (DGX Spark): trtllm SM100-only, tilelang exceeds dynamic smem, triton ROCm-only, flashinfer_sparse_mla gated to GlmMoeDsa archs
- [Bug] FlashInfer autotune cache is discarded every boot under MoE expert parallelism (EP>1): per-rank shapes never agree, so the fused-MoE tactic is re-drawn each start
- Docs
- Python not yet supported